Re: [Mifluz-dev] Question on API

[Brian Aker]

On Sun, 2002-02-17 at 20:43, Geoff Hutchison wrote:
> > 7) Needs to be able to search and index that was created with 1 gig of
> > text in under 3 seconds.
> 
> No offense, but this is a bit nonsensical. Granted, I'll assume that 
Woops, you are right.

> you're going to back things up with reliable, fast hardware. But there's 
> a great deal of difference between say, 1 billion keys with a few bytes 
> of record attached and 1 million keys with a few K of record. Things 
> generally scale by the number of keys more than anything else. Even so, 
> unless you're return a lot of query hits and need to do significant work 
> before presentation, 3 seconds is a lot of CPU time.
What I am looking at is Slashdot's comments. So we are talking about 2.3
million comments with around 7000 entries added per day. Three seconds
would be about the maximum amount of time I would want to use for the
query (page generation and other actions require some time too). So the
entire effort needs to stay under 5 seconds.

> > 3) Allows me to pass it a query string and have it return the unique
> > keyword that the text was identified with when it was inserted (and it
> > would be really nice if it gave back some sort of number representing
> > the matched value).
> 
> I'm not quite sure I follow. This sounds like you want the query to 
> match the blob (i.e. the record) and return the key? Normally you'd use 
> the keywords to retrieve the blob. Or am I misunderstanding you?
So I have a comment id which is a number, I would need to insert the
comment blob and when I make a query it would need to return to me the
the comment id, and a number that I could use to sort the returns.
        -Brian


-- 
_______________________________________________________
Brian Aker, address@hidden 
Slashdot Senior Developer
Seattle, Washington
http://tangent.org/~brian/
http://askbrian.org/
_______________________________________________________
You can't grep a dead tree.